surya.narada.systems

R6: system architecture as research, mapped

Sweep executed 19 August 2026. The finding: R6's four sub-directions sit in four different evidence classes. High-voltage arrays are established heritage with a known physical constraint. The Starlink standardization lesson is confirmed by independent analyses. The qualification moat is now priced. And the headline claim — ISCR's ~500 W/kg by merging solar, radiator, and compute into one structure — rests on a single unrefereed preprint with no hardware behind it.

Why this map exists

R6 is the cross-cutting direction: cheap to do now, compounding returns. The directions page asserts the ISCR move (~500 W/kg at array level), the high-voltage lever, the standardization lesson, and qualification economics as an attack on the filing-cabinet moat. This page checks each assertion's evidence class — because a program that quotes 500 W/kg needs to know whether it is quoting a measurement or a whiteboard.

Sub-direction inventory

Sub-directionBest evidenceWhat's missingEvidence class
ISCR integration (solar + radiator + compute)One unrefereed preprint (Gaalema, Indyk, Staley, arXiv April 2026): co-locating cells, vapor-chamber radiator, and compute on common support yields ~500 W/kg at array level and >100 kW of compute per launched tonne; a 16 MW / 150 t / 20 m × 2,200 m grid of 16,000 panels fits one Starship. One commercial parallel: Sophia Space's radiator-area orbital data center bet (patent-backed).Independent replication, any hardware, thermal-vacuum test of a panel, deployment dynamics of a kilometer-class gridsingle-source
High-voltage arraysEstablished heritage: ISS runs 160 V primary power, a decision made explicitly to cut cable mass; direct-drive SEP concepts run 300–400 V because a 400 kW vehicle at 100–160 V must distribute up to 4,000 A — cabling becomes prohibitive. Test campaigns exist (D2HET high-voltage array arc testing; EDDA H2020 direct-drive hardware results).The constraint is known and physical: arcing in LEO plasma. High voltage lives at high power SEP / GEO class, or behind plasma mitigation — not a free win at LEO constellation altitudeestablished, regime-bounded
Mass-production panel standardizationThe Starlink lesson, now confirmed by independent analyses: SemiAnalysis (2026) finds silicon's mass and cost advantage at Starlink-class production outweighs GaAs efficiency for 30 kW–1 MW compute payloads; Ex Terra (2026): GaAs was prohibitively expensive at Starlink volume. Combined with the R5 sweep anchors: COTS Si at $0.20–0.50/W vs 3J at $50–100/W, ~70 satellites/week in Redmond.A published teardown of Starlink panel cost structure — everything is inference from production economics, not disclosed BOMconfirmed directionally
Qualification economicsThe moat is priced: a full AIAA S-111A + S-112A campaign ran ~$4.5M and 1–2+ years (STRATFI award record, from the R2 sweep). Beam time is ~$800/hour with most cost in test-bench development, and ~33% of facility requests go unsatisfied (CERN RADWG/RCS data). GSFC-STD-1000H mandates S-111/S-112 by rule; incumbents sell qualification as the product (Rocket Lab: 'fully qualified to AIAA-S111-2014').No published low-cost qualification protocol; no pooled-facility model; nobody sells a fast campaignmoat confirmed, attack unbuilt

The ISCR claim, graded

single-source The ~500 W/kg figure is real arithmetic in a real preprint, and the physical intuition is sound — a radiator's structure can carry cells, and a 35 °C junction beats a 105 °C one on energy per token by >30%. But the entire evidence base is one April 2026 arXiv preprint (Gaalema, Indyk, Staley), explicitly not peer reviewed, plus one startup's patent-backed bet on the same radiator-area thesis (Sophia Space). No panel has been built, no thermal-vacuum test published, no deployment dynamics analyzed for a 20 m × 2,200 m grid. Treat the number as a design target with a plausible mechanism, not as demonstrated specific power.

What would move ISCR up an evidence class: a single benchtop panel — cells bonded to a vapor-chamber radiator with compute on the back, run through thermal-vacuum with IV monitoring. That is a ground campaign at an existing facility, same shape as the R2 datasets. It is also exactly the kind of artifact that forces the thermal interfaces, CTE stack, and harness routing into the open, where the real mass accounting lives.

High voltage: established, with a fence

established The lever is heritage, not speculation. ISS chose 160 V primary power explicitly for cable mass. Direct-drive SEP studies put the array at thruster voltage (300–400 V) because a 400 kW vehicle at conventional voltage distributes up to 4,000 A — the harness becomes the mass problem. The fence is equally well documented: high-voltage arrays arc in LEO plasma, and the mitigation literature (arc testing, grouting, plasma contactors) is its own field. High voltage pays at high-power SEP and GEO class; at LEO constellation altitude it buys a plasma problem. Regime-bounded, same shape as the R5 atomic-oxygen fence.

Standardization: the lesson holds up

confirmed The Starlink lesson survives independent scrutiny. SemiAnalysis' 2026 space-datacenter case finds silicon's mass and cost advantage at Starlink-class production outweighs GaAs efficiency for 30 kW–1 MW payloads; Ex Terra's account matches: GaAs was prohibitively expensive at Starlink volume. Combined with the R5 sweep's anchors (COTS Si at $0.20–0.50/W against 3J at $50–100/W; ~70 satellites/week), the conclusion hardens: at constellation economics, cell choice is a second-order decision behind panel standardization and production cadence. The honest perimeter: no published Starlink panel BOM teardown exists — this is inference from production economics, consistently pointing one way.

Qualification economics: the moat has a price list

established The R2 sweep priced the campaign (~$4.5M, 1–2+ years for S-111A + S-112A via the STRATFI record). This sweep priced the inputs: beam time ~$800/hour with most cost in test-bench development, and ~33% of radiation-facility requests unsatisfied — the queue is part of the moat. The mandate is codified (GSFC-STD-1000H requires S-111/S-112 by rule), and the incumbent sells the moat as the product: Rocket Lab's cell page leads with "fully qualified to AIAA-S111-2014." Nobody publishes a low-cost protocol, nobody pools facility access, nobody sells a fast campaign. The attack the directions page sketched — cheap, fast radiation-test campaigns that let new materials accumulate credible heritage — remains unbuilt by anyone.

What this changes

Program moves, in order: (1) quote ISCR as a design target, never as a demonstrated number — and consider the benchtop panel above as the cheapest high-information artifact in the whole portfolio; (2) treat high voltage as a regime tool for the R5/GEO-class concepts, not a LEO play; (3) keep standardization as the default cost model assumption — the burden of proof now sits on any architecture that needs exotic cells at volume; (4) scope the fast-qualification campaign as a deliverable: the moat is priced, the queue is documented, and the counter-position (pooled facility + fixed-price campaign) is unclaimed. R6 earns its "compounding returns" label — every other direction's numbers get sharper when the architecture assumptions are graded.

Perimeter and provenance

Four-track sweep (ISCR, high-voltage heritage, standardization, qualification economics), 12 new sources, admitted to the Surya epistemic graph as event ev-000000000012 with assessment surya-assessment-r6-gap-map-2026-08, judgment established-with-gap. Sweep-level gaps, flagged honestly: the ISCR preprint was assessed from its abstract and coverage, not a full-text read; SemiAnalysis and Ex Terra are analyst pieces, not primary disclosures; the STRATFI $4.5M anchor is carried from the R2 sweep (event ev-000000000007), not re-verified; no Starlink BOM teardown was located. None of these move a sub-direction out of its evidence class.

Sources